GETTING STARTED

AlphaFold3 Mac Port overview¶

Run AlphaFold 3 on Apple Silicon with an MLX-native inference stack while keeping AF3-compatible data pipeline behavior.

Get started in 10 minutes¶

Prerequisites:

Apple Silicon Mac (M2/M3/M4 recommended)
Python 3.12 environment
AF3 model weights in weights/model/af3.bin.zst

Install and run:

source .venv/bin/activate
PYTHONPATH=src python3 run_alphafold_mlx.py \
  --input examples/desi1_monomer.json \
  --output_dir output/quickstart \
  --num_samples 1 \
  --diffusion_steps 20 \
  --verbose

Info

Continue with Quickstart for the full walkthrough and expected outputs.

Warning

AlphaFold 3 weights are required for real inference and are subject to usage terms. Review the weights terms before running production workloads.

Where to go next¶

Installation for complete machine and dependency setup
Pipeline Setup (macOS) for full AF3-style MSA/templates
Input Format and Output Format
Web Interface Guide for all UI flows and controls
Troubleshooting for common failures
CLI Reference for all runtime flags
API Reference for all REST and WebSocket endpoints

Project context¶

This repository is a macOS-focused port of AlphaFold 3 inference using MLX while preserving AF3 data and output contracts.

For the upstream project and legal terms, see AlphaFold 3 upstream.